4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 #include <sys/types.h>
27 #include <sys/param.h>
28 #include <sys/systm.h>
29 #include <sys/kernel.h>
30 #include <sys/sysmacros.h>
33 #include <sys/vnode.h>
35 #include <sys/mntent.h>
36 #include <sys/mount.h>
37 #include <sys/cmn_err.h>
38 #include <sys/zfs_znode.h>
39 #include <sys/zfs_dir.h>
41 #include <sys/fs/zfs.h>
43 #include <sys/dsl_prop.h>
44 #include <sys/dsl_dataset.h>
45 #include <sys/dsl_deleg.h>
48 #include <sys/varargs.h>
49 #include <sys/policy.h>
50 #include <sys/atomic.h>
51 #include <sys/zfs_ioctl.h>
52 #include <sys/zfs_ctldir.h>
53 #include <sys/zfs_fuid.h>
54 #include <sys/sunddi.h>
56 #include <sys/dmu_objset.h>
57 #include <sys/spa_boot.h>
58 #include <sys/vdev_impl.h> /* VDEV_BOOT_VERSION */
60 struct mtx zfs_debug_mtx;
61 MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF);
63 SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system");
65 int zfs_super_owner = 0;
66 SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0,
67 "File system owner can perform privileged operation on his file systems");
69 int zfs_debug_level = 0;
70 TUNABLE_INT("vfs.zfs.debug", &zfs_debug_level);
71 SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RW, &zfs_debug_level, 0,
74 SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions");
75 static int zfs_version_acl = ZFS_ACL_VERSION;
76 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0,
78 static int zfs_version_dmu_backup_header = DMU_BACKUP_HEADER_VERSION;
79 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, dmu_backup_header, CTLFLAG_RD,
80 &zfs_version_dmu_backup_header, 0, "DMU_BACKUP_HEADER_VERSION");
81 static int zfs_version_dmu_backup_stream = DMU_BACKUP_STREAM_VERSION;
82 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, dmu_backup_stream, CTLFLAG_RD,
83 &zfs_version_dmu_backup_stream, 0, "DMU_BACKUP_STREAM_VERSION");
84 static int zfs_version_spa = SPA_VERSION;
85 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0,
87 static int zfs_version_vdev_boot = VDEV_BOOT_VERSION;
88 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, vdev_boot, CTLFLAG_RD,
89 &zfs_version_vdev_boot, 0, "VDEV_BOOT_VERSION");
90 static int zfs_version_zpl = ZPL_VERSION;
91 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0,
94 static int zfs_mount(vfs_t *vfsp);
95 static int zfs_umount(vfs_t *vfsp, int fflag);
96 static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp);
97 static int zfs_statfs(vfs_t *vfsp, struct statfs *statp);
98 static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp);
99 static int zfs_sync(vfs_t *vfsp, int waitfor);
100 static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
101 struct ucred **credanonp, int *numsecflavors, int **secflavors);
102 static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, vnode_t **vpp);
103 static void zfs_objset_close(zfsvfs_t *zfsvfs);
104 static void zfs_freevfs(vfs_t *vfsp);
106 static struct vfsops zfs_vfsops = {
107 .vfs_mount = zfs_mount,
108 .vfs_unmount = zfs_umount,
109 .vfs_root = zfs_root,
110 .vfs_statfs = zfs_statfs,
111 .vfs_vget = zfs_vget,
112 .vfs_sync = zfs_sync,
113 .vfs_checkexp = zfs_checkexp,
114 .vfs_fhtovp = zfs_fhtovp,
117 VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN);
120 * We need to keep a count of active fs's.
121 * This is necessary to prevent our module
122 * from being unloaded after a umount -f
124 static uint32_t zfs_active_fs_count = 0;
128 zfs_sync(vfs_t *vfsp, int waitfor)
132 * Data integrity is job one. We don't want a compromised kernel
133 * writing to the storage pool, so we never sync during panic.
140 * Sync a specific filesystem.
142 zfsvfs_t *zfsvfs = vfsp->vfs_data;
145 error = vfs_stdsync(vfsp, waitfor);
150 if (zfsvfs->z_log != NULL)
151 zil_commit(zfsvfs->z_log, UINT64_MAX, 0);
153 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
157 * Sync all ZFS filesystems. This is what happens when you
158 * run sync(1M). Unlike other filesystems, ZFS honors the
159 * request by waiting for all pools to commit all dirty data.
168 atime_changed_cb(void *arg, uint64_t newval)
170 zfsvfs_t *zfsvfs = arg;
172 if (newval == TRUE) {
173 zfsvfs->z_atime = TRUE;
174 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME;
175 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
176 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
178 zfsvfs->z_atime = FALSE;
179 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME;
180 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
181 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
186 xattr_changed_cb(void *arg, uint64_t newval)
188 zfsvfs_t *zfsvfs = arg;
190 if (newval == TRUE) {
191 /* XXX locking on vfs_flag? */
193 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
195 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
196 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
198 /* XXX locking on vfs_flag? */
200 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
202 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
203 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
208 blksz_changed_cb(void *arg, uint64_t newval)
210 zfsvfs_t *zfsvfs = arg;
212 if (newval < SPA_MINBLOCKSIZE ||
213 newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
214 newval = SPA_MAXBLOCKSIZE;
216 zfsvfs->z_max_blksz = newval;
217 zfsvfs->z_vfs->vfs_bsize = newval;
221 readonly_changed_cb(void *arg, uint64_t newval)
223 zfsvfs_t *zfsvfs = arg;
226 /* XXX locking on vfs_flag? */
227 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
228 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
229 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
231 /* XXX locking on vfs_flag? */
232 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
233 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
234 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
239 setuid_changed_cb(void *arg, uint64_t newval)
241 zfsvfs_t *zfsvfs = arg;
243 if (newval == FALSE) {
244 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
245 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
246 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
248 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
249 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
250 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
255 exec_changed_cb(void *arg, uint64_t newval)
257 zfsvfs_t *zfsvfs = arg;
259 if (newval == FALSE) {
260 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
261 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
262 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
264 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
265 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
266 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
271 * The nbmand mount option can be changed at mount time.
272 * We can't allow it to be toggled on live file systems or incorrect
273 * behavior may be seen from cifs clients
275 * This property isn't registered via dsl_prop_register(), but this callback
276 * will be called when a file system is first mounted
279 nbmand_changed_cb(void *arg, uint64_t newval)
281 zfsvfs_t *zfsvfs = arg;
282 if (newval == FALSE) {
283 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
284 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
286 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
287 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
292 snapdir_changed_cb(void *arg, uint64_t newval)
294 zfsvfs_t *zfsvfs = arg;
296 zfsvfs->z_show_ctldir = newval;
300 vscan_changed_cb(void *arg, uint64_t newval)
302 zfsvfs_t *zfsvfs = arg;
304 zfsvfs->z_vscan = newval;
308 acl_mode_changed_cb(void *arg, uint64_t newval)
310 zfsvfs_t *zfsvfs = arg;
312 zfsvfs->z_acl_mode = newval;
316 acl_inherit_changed_cb(void *arg, uint64_t newval)
318 zfsvfs_t *zfsvfs = arg;
320 zfsvfs->z_acl_inherit = newval;
324 zfs_register_callbacks(vfs_t *vfsp)
326 struct dsl_dataset *ds = NULL;
328 zfsvfs_t *zfsvfs = NULL;
330 int readonly, do_readonly = FALSE;
331 int setuid, do_setuid = FALSE;
332 int exec, do_exec = FALSE;
333 int xattr, do_xattr = FALSE;
334 int atime, do_atime = FALSE;
338 zfsvfs = vfsp->vfs_data;
343 * This function can be called for a snapshot when we update snapshot's
344 * mount point, which isn't really supported.
346 if (dmu_objset_is_snapshot(os))
350 * The act of registering our callbacks will destroy any mount
351 * options we may have. In order to enable temporary overrides
352 * of mount options, we stash away the current values and
353 * restore them after we register the callbacks.
355 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL)) {
357 do_readonly = B_TRUE;
358 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
360 do_readonly = B_TRUE;
362 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
366 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
369 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
374 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
377 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
381 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
384 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
388 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
391 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
397 * nbmand is a special property. It can only be changed at
400 * This is weird, but it is documented to only be changeable
403 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
405 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
408 char osname[MAXNAMELEN];
410 dmu_objset_name(os, osname);
411 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
418 * Register property callbacks.
420 * It would probably be fine to just check for i/o error from
421 * the first prop_register(), but I guess I like to go
424 ds = dmu_objset_ds(os);
425 error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs);
426 error = error ? error : dsl_prop_register(ds,
427 "xattr", xattr_changed_cb, zfsvfs);
428 error = error ? error : dsl_prop_register(ds,
429 "recordsize", blksz_changed_cb, zfsvfs);
430 error = error ? error : dsl_prop_register(ds,
431 "readonly", readonly_changed_cb, zfsvfs);
432 error = error ? error : dsl_prop_register(ds,
433 "setuid", setuid_changed_cb, zfsvfs);
434 error = error ? error : dsl_prop_register(ds,
435 "exec", exec_changed_cb, zfsvfs);
436 error = error ? error : dsl_prop_register(ds,
437 "snapdir", snapdir_changed_cb, zfsvfs);
438 error = error ? error : dsl_prop_register(ds,
439 "aclmode", acl_mode_changed_cb, zfsvfs);
440 error = error ? error : dsl_prop_register(ds,
441 "aclinherit", acl_inherit_changed_cb, zfsvfs);
442 error = error ? error : dsl_prop_register(ds,
443 "vscan", vscan_changed_cb, zfsvfs);
448 * Invoke our callbacks to restore temporary mount options.
451 readonly_changed_cb(zfsvfs, readonly);
453 setuid_changed_cb(zfsvfs, setuid);
455 exec_changed_cb(zfsvfs, exec);
457 xattr_changed_cb(zfsvfs, xattr);
459 atime_changed_cb(zfsvfs, atime);
461 nbmand_changed_cb(zfsvfs, nbmand);
467 * We may attempt to unregister some callbacks that are not
468 * registered, but this is OK; it will simply return ENOMSG,
469 * which we will ignore.
471 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs);
472 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs);
473 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs);
474 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs);
475 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs);
476 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs);
477 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs);
478 (void) dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, zfsvfs);
479 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
481 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs);
487 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
491 error = zfs_register_callbacks(zfsvfs->z_vfs);
496 * Set the objset user_ptr to track its zfsvfs.
498 mutex_enter(&zfsvfs->z_os->os->os_user_ptr_lock);
499 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
500 mutex_exit(&zfsvfs->z_os->os->os_user_ptr_lock);
503 * If we are not mounting (ie: online recv), then we don't
504 * have to worry about replaying the log as we blocked all
505 * operations out since we closed the ZIL.
511 * During replay we remove the read only flag to
512 * allow replays to succeed.
514 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
515 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
518 * Parse and replay the intent log.
520 zil_replay(zfsvfs->z_os, zfsvfs, &zfsvfs->z_assign,
521 zfs_replay_vector, zfs_unlinked_drain);
523 zfs_unlinked_drain(zfsvfs);
524 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
528 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
534 zfs_freezfsvfs(zfsvfs_t *zfsvfs)
536 mutex_destroy(&zfsvfs->z_znodes_lock);
537 mutex_destroy(&zfsvfs->z_online_recv_lock);
538 list_destroy(&zfsvfs->z_all_znodes);
539 rrw_destroy(&zfsvfs->z_teardown_lock);
540 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
541 rw_destroy(&zfsvfs->z_fuid_lock);
542 kmem_free(zfsvfs, sizeof (zfsvfs_t));
546 zfs_domount(vfs_t *vfsp, char *osname)
548 uint64_t recordsize, readonly;
558 * Initialize the zfs-specific filesystem structure.
559 * Should probably make this a kmem cache, shuffle fields,
560 * and just bzero up to z_hold_mtx[].
562 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
563 zfsvfs->z_vfs = vfsp;
564 zfsvfs->z_parent = zfsvfs;
565 zfsvfs->z_assign = TXG_NOWAIT;
566 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
567 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
569 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
570 mutex_init(&zfsvfs->z_online_recv_lock, NULL, MUTEX_DEFAULT, NULL);
571 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
572 offsetof(znode_t, z_link_node));
573 rrw_init(&zfsvfs->z_teardown_lock);
574 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
575 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
577 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
580 zfsvfs->z_vfs->vfs_bsize = recordsize;
582 vfsp->vfs_data = zfsvfs;
583 vfsp->mnt_flag |= MNT_LOCAL;
584 vfsp->mnt_kern_flag |= MNTK_MPSAFE;
585 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
586 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
588 if (error = dsl_prop_get_integer(osname, "readonly", &readonly, NULL))
591 mode = DS_MODE_OWNER;
593 mode |= DS_MODE_READONLY;
595 error = dmu_objset_open(osname, DMU_OST_ZFS, mode, &zfsvfs->z_os);
596 if (error == EROFS) {
597 mode = DS_MODE_OWNER | DS_MODE_READONLY;
598 error = dmu_objset_open(osname, DMU_OST_ZFS, mode,
605 if (error = zfs_init_fs(zfsvfs, &zp))
609 * Set features for file system.
611 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
612 if (zfsvfs->z_use_fuids) {
613 vfs_set_feature(vfsp, VFSFT_XVATTR);
614 vfs_set_feature(vfsp, VFSFT_SYSATTR_VIEWS);
615 vfs_set_feature(vfsp, VFSFT_ACEMASKONACCESS);
616 vfs_set_feature(vfsp, VFSFT_ACLONCREATE);
618 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
619 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
620 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
621 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
622 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
623 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
624 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
627 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
630 ASSERT(mode & DS_MODE_READONLY);
631 atime_changed_cb(zfsvfs, B_FALSE);
632 readonly_changed_cb(zfsvfs, B_TRUE);
633 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
635 xattr_changed_cb(zfsvfs, pval);
636 zfsvfs->z_issnap = B_TRUE;
638 error = zfsvfs_setup(zfsvfs, B_TRUE);
641 vfs_mountedfrom(vfsp, osname);
643 if (!zfsvfs->z_issnap)
644 zfsctl_create(zfsvfs);
648 dmu_objset_close(zfsvfs->z_os);
649 zfs_freezfsvfs(zfsvfs);
651 atomic_add_32(&zfs_active_fs_count, 1);
658 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
660 objset_t *os = zfsvfs->z_os;
661 struct dsl_dataset *ds;
664 * Unregister properties.
666 if (!dmu_objset_is_snapshot(os)) {
667 ds = dmu_objset_ds(os);
668 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
671 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
674 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
677 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
680 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
683 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
686 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
689 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb,
692 VERIFY(dsl_prop_unregister(ds, "aclinherit",
693 acl_inherit_changed_cb, zfsvfs) == 0);
695 VERIFY(dsl_prop_unregister(ds, "vscan",
696 vscan_changed_cb, zfsvfs) == 0);
702 zfs_mount(vfs_t *vfsp)
704 kthread_t *td = curthread;
705 vnode_t *mvp = vfsp->mnt_vnodecovered;
706 cred_t *cr = td->td_ucred;
711 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
715 * If full-owner-access is enabled and delegated administration is
716 * turned on, we must set nosuid.
718 if (zfs_super_owner &&
719 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
720 secpolicy_fs_mount_clearopts(cr, vfsp);
724 * Check for mount privilege?
726 * If we don't have privilege then see if
727 * we have local permission to allow it
729 error = secpolicy_fs_mount(cr, mvp, vfsp);
731 error = dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr);
735 if (!(vfsp->vfs_flag & MS_REMOUNT)) {
739 * Make sure user is the owner of the mount point
740 * or has sufficient privileges.
743 vattr.va_mask = AT_UID;
745 vn_lock(mvp, LK_SHARED | LK_RETRY);
746 if (error = VOP_GETATTR(mvp, &vattr, cr)) {
751 #if 0 /* CHECK THIS! Is probably needed for zfs_suser. */
752 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
753 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
758 if (error = secpolicy_vnode_owner(mvp, cr, vattr.va_uid)) {
763 if (error = VOP_ACCESS(mvp, VWRITE, cr, td)) {
771 secpolicy_fs_mount_clearopts(cr, vfsp);
775 * Refuse to mount a filesystem if we are in a local zone and the
776 * dataset is not visible.
778 if (!INGLOBALZONE(curthread) &&
779 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
785 * When doing a remount, we simply refresh our temporary properties
786 * according to those options set in the current VFS options.
788 if (vfsp->vfs_flag & MS_REMOUNT) {
789 /* refresh mount options */
790 zfs_unregister_callbacks(vfsp->vfs_data);
791 error = zfs_register_callbacks(vfsp);
796 error = zfs_domount(vfsp, osname);
803 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
805 zfsvfs_t *zfsvfs = vfsp->vfs_data;
806 uint64_t refdbytes, availbytes, usedobjs, availobjs;
808 statp->f_version = STATFS_VERSION;
812 dmu_objset_space(zfsvfs->z_os,
813 &refdbytes, &availbytes, &usedobjs, &availobjs);
816 * The underlying storage pool actually uses multiple block sizes.
817 * We report the fragsize as the smallest block size we support,
818 * and we report our blocksize as the filesystem's maximum blocksize.
820 statp->f_bsize = zfsvfs->z_vfs->vfs_bsize;
821 statp->f_iosize = zfsvfs->z_vfs->vfs_bsize;
824 * The following report "total" blocks of various kinds in the
825 * file system, but reported in terms of f_frsize - the
829 statp->f_blocks = (refdbytes + availbytes) / statp->f_bsize;
830 statp->f_bfree = availbytes / statp->f_bsize;
831 statp->f_bavail = statp->f_bfree; /* no root reservation */
834 * statvfs() should really be called statufs(), because it assumes
835 * static metadata. ZFS doesn't preallocate files, so the best
836 * we can do is report the max that could possibly fit in f_files,
837 * and that minus the number actually used in f_ffree.
838 * For f_ffree, report the smaller of the number of object available
839 * and the number of blocks (each object will take at least a block).
841 statp->f_ffree = MIN(availobjs, statp->f_bfree);
842 statp->f_files = statp->f_ffree + usedobjs;
845 * We're a zfs filesystem.
847 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));
849 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
850 sizeof(statp->f_mntfromname));
851 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
852 sizeof(statp->f_mntonname));
854 statp->f_namemax = ZFS_MAXNAMELEN;
861 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
863 zfsvfs_t *zfsvfs = vfsp->vfs_data;
869 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
872 error = vn_lock(*vpp, flags);
873 (*vpp)->v_vflag |= VV_ROOT;
881 * Teardown the zfsvfs::z_os.
883 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
884 * and 'z_teardown_inactive_lock' held.
887 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
891 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
895 * We purge the parent filesystem's vfsp as the parent
896 * filesystem and all of its snapshots have their vnode's
897 * v_vfsp set to the parent's filesystem's vfsp. Note,
898 * 'z_parent' is self referential for non-snapshots.
900 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
904 * Close the zil. NB: Can't close the zil while zfs_inactive
905 * threads are blocked as zil_close can call zfs_inactive.
908 zil_close(zfsvfs->z_log);
909 zfsvfs->z_log = NULL;
912 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
915 * If we are not unmounting (ie: online recv) and someone already
916 * unmounted this file system while we were doing the switcheroo,
917 * or a reopen of z_os failed then just bail out now.
919 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
920 rw_exit(&zfsvfs->z_teardown_inactive_lock);
921 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
926 * At this point there are no vops active, and any new vops will
927 * fail with EIO since we have z_teardown_lock for writer (only
928 * relavent for forced unmount).
930 * Release all holds on dbufs.
932 mutex_enter(&zfsvfs->z_znodes_lock);
933 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
934 zp = list_next(&zfsvfs->z_all_znodes, zp))
936 ASSERT(ZTOV(zp)->v_count >= 0);
937 zfs_znode_dmu_fini(zp);
939 mutex_exit(&zfsvfs->z_znodes_lock);
942 * If we are unmounting, set the unmounted flag and let new vops
943 * unblock. zfs_inactive will have the unmounted behavior, and all
944 * other vops will fail with EIO.
947 zfsvfs->z_unmounted = B_TRUE;
948 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
949 rw_exit(&zfsvfs->z_teardown_inactive_lock);
953 * Some znodes might not be fully reclaimed, wait for them.
955 mutex_enter(&zfsvfs->z_znodes_lock);
956 while (list_head(&zfsvfs->z_all_znodes) != NULL) {
957 msleep(zfsvfs, &zfsvfs->z_znodes_lock, 0,
960 mutex_exit(&zfsvfs->z_znodes_lock);
965 * z_os will be NULL if there was an error in attempting to reopen
966 * zfsvfs, so just return as the properties had already been
967 * unregistered and cached data had been evicted before.
969 if (zfsvfs->z_os == NULL)
973 * Unregister properties.
975 zfs_unregister_callbacks(zfsvfs);
980 if (dmu_objset_evict_dbufs(zfsvfs->z_os)) {
981 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
982 (void) dmu_objset_evict_dbufs(zfsvfs->z_os);
990 zfs_umount(vfs_t *vfsp, int fflag)
992 zfsvfs_t *zfsvfs = vfsp->vfs_data;
994 cred_t *cr = curthread->td_ucred;
997 if (fflag & MS_FORCE) {
998 /* TODO: Force unmount is not well implemented yet, so deny it. */
999 ZFS_LOG(0, "Force unmount is experimental - report any problems.");
1002 ret = secpolicy_fs_unmount(cr, vfsp);
1004 ret = dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1005 ZFS_DELEG_PERM_MOUNT, cr);
1010 * We purge the parent filesystem's vfsp as the parent filesystem
1011 * and all of its snapshots have their vnode's v_vfsp set to the
1012 * parent's filesystem's vfsp. Note, 'z_parent' is self
1013 * referential for non-snapshots.
1015 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1018 * Unmount any snapshots mounted under .zfs before unmounting the
1021 if (zfsvfs->z_ctldir != NULL) {
1022 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
1024 ret = vflush(vfsp, 0, 0, curthread);
1025 ASSERT(ret == EBUSY);
1026 if (!(fflag & MS_FORCE)) {
1027 if (zfsvfs->z_ctldir->v_count > 1)
1029 ASSERT(zfsvfs->z_ctldir->v_count == 1);
1031 zfsctl_destroy(zfsvfs);
1032 ASSERT(zfsvfs->z_ctldir == NULL);
1036 * Flush all the files.
1038 ret = vflush(vfsp, 1, (fflag & MS_FORCE) ? FORCECLOSE : 0, curthread);
1040 if (!zfsvfs->z_issnap) {
1041 zfsctl_create(zfsvfs);
1042 ASSERT(zfsvfs->z_ctldir != NULL);
1047 if (!(fflag & MS_FORCE)) {
1049 * Check the number of active vnodes in the file system.
1050 * Our count is maintained in the vfs structure, but the
1051 * number is off by 1 to indicate a hold on the vfs
1054 * The '.zfs' directory maintains a reference of its
1055 * own, and any active references underneath are
1056 * reflected in the vnode count.
1058 if (zfsvfs->z_ctldir == NULL) {
1059 if (vfsp->vfs_count > 1)
1062 if (vfsp->vfs_count > 2 ||
1063 zfsvfs->z_ctldir->v_count > 1)
1068 vfsp->mnt_kern_flag |= MNTK_UNMOUNTF;
1072 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1076 * z_os will be NULL if there was an error in
1077 * attempting to reopen zfsvfs.
1081 * Unset the objset user_ptr.
1083 mutex_enter(&os->os->os_user_ptr_lock);
1084 dmu_objset_set_user(os, NULL);
1085 mutex_exit(&os->os->os_user_ptr_lock);
1088 * Finally release the objset
1090 dmu_objset_close(os);
1094 * We can now safely destroy the '.zfs' directory node.
1096 if (zfsvfs->z_ctldir != NULL)
1097 zfsctl_destroy(zfsvfs);
1098 if (zfsvfs->z_issnap) {
1099 vnode_t *svp = vfsp->mnt_vnodecovered;
1101 ASSERT(svp->v_count == 2 || svp->v_count == 1);
1102 if (svp->v_count == 2)
1111 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
1113 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1118 * XXXPJD: zfs_zget() can't operate on virtual entires like .zfs/ or
1119 * .zfs/snapshot/ directories, so for now just return EOPNOTSUPP.
1120 * This will make NFS to fall back to using READDIR instead of
1122 * Also snapshots are stored in AVL tree, but based on their names,
1123 * not inode numbers, so it will be very inefficient to iterate
1124 * over all snapshots to find the right one.
1125 * Note that OpenSolaris READDIRPLUS implementation does LOOKUP on
1126 * d_name, and not VGET on d_fileno as we do.
1128 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR)
1129 return (EOPNOTSUPP);
1132 err = zfs_zget(zfsvfs, ino, &zp);
1133 if (err == 0 && zp->z_unlinked) {
1141 vn_lock(*vpp, flags);
1148 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
1149 struct ucred **credanonp, int *numsecflavors, int **secflavors)
1151 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1154 * If this is regular file system vfsp is the same as
1155 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
1156 * zfsvfs->z_parent->z_vfs represents parent file system
1157 * which we have to use here, because only this file system
1158 * has mnt_export configured.
1160 vfsp = zfsvfs->z_parent->z_vfs;
1162 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
1163 credanonp, numsecflavors, secflavors));
1166 CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
1167 CTASSERT(LONG_FID_LEN <= sizeof(struct fid));
1170 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, vnode_t **vpp)
1172 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1174 uint64_t object = 0;
1175 uint64_t fid_gen = 0;
1185 * On FreeBSD we can get snapshot's mount point or its parent file
1186 * system mount point depending if snapshot is already mounted or not.
1188 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
1189 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1190 uint64_t objsetid = 0;
1191 uint64_t setgen = 0;
1193 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1194 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1196 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1197 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1201 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1207 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1208 zfid_short_t *zfid = (zfid_short_t *)fidp;
1210 for (i = 0; i < sizeof (zfid->zf_object); i++)
1211 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1213 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1214 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1220 /* A zero fid_gen means we are in the .zfs control directories */
1222 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1223 *vpp = zfsvfs->z_ctldir;
1224 ASSERT(*vpp != NULL);
1225 if (object == ZFSCTL_INO_SNAPDIR) {
1226 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
1227 0, NULL, NULL, NULL, NULL, NULL) == 0);
1231 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
1236 gen_mask = -1ULL >> (64 - 8 * i);
1238 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1239 if (err = zfs_zget(zfsvfs, object, &zp)) {
1243 zp_gen = zp->z_phys->zp_gen & gen_mask;
1246 if (zp->z_unlinked || zp_gen != fid_gen) {
1247 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
1254 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
1255 vnode_create_vobject(*vpp, zp->z_phys->zp_size, curthread);
1261 * Block out VOPs and close zfsvfs_t::z_os
1263 * Note, if successful, then we return with the 'z_teardown_lock' and
1264 * 'z_teardown_inactive_lock' write held.
1267 zfs_suspend_fs(zfsvfs_t *zfsvfs, char *name, int *mode)
1271 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
1274 *mode = zfsvfs->z_os->os_mode;
1275 dmu_objset_name(zfsvfs->z_os, name);
1276 dmu_objset_close(zfsvfs->z_os);
1282 * Reopen zfsvfs_t::z_os and release VOPs.
1285 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname, int mode)
1289 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
1290 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1292 err = dmu_objset_open(osname, DMU_OST_ZFS, mode, &zfsvfs->z_os);
1294 zfsvfs->z_os = NULL;
1298 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
1301 * Attempt to re-establish all the active znodes with
1302 * their dbufs. If a zfs_rezget() fails, then we'll let
1303 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
1304 * when they try to use their znode.
1306 mutex_enter(&zfsvfs->z_znodes_lock);
1307 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
1308 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1309 (void) zfs_rezget(zp);
1311 mutex_exit(&zfsvfs->z_znodes_lock);
1315 /* release the VOPs */
1316 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1317 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1321 * Since we couldn't reopen zfsvfs::z_os, force
1322 * unmount this file system.
1324 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
1325 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
1331 zfs_freevfs(vfs_t *vfsp)
1333 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1336 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1337 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1339 zfs_fuid_destroy(zfsvfs);
1340 zfs_freezfsvfs(zfsvfs);
1342 atomic_add_32(&zfs_active_fs_count, -1);
1346 static int desiredvnodes_backup;
1350 zfs_vnodes_adjust(void)
1353 int newdesiredvnodes;
1355 desiredvnodes_backup = desiredvnodes;
1358 * We calculate newdesiredvnodes the same way it is done in
1359 * vntblinit(). If it is equal to desiredvnodes, it means that
1360 * it wasn't tuned by the administrator and we can tune it down.
1362 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 *
1363 vm_kmem_size / (5 * (sizeof(struct vm_object) +
1364 sizeof(struct vnode))));
1365 if (newdesiredvnodes == desiredvnodes)
1366 desiredvnodes = (3 * newdesiredvnodes) / 4;
1371 zfs_vnodes_adjust_back(void)
1375 desiredvnodes = desiredvnodes_backup;
1383 printf("ZFS filesystem version " SPA_VERSION_STRING "\n");
1386 * Initialize znode cache, vnode ops, etc...
1391 * Initialize .zfs directory structures
1396 * Reduce number of vnode. Originally number of vnodes is calculated
1397 * with UFS inode in mind. We reduce it here, because it's too big for
1400 zfs_vnodes_adjust();
1408 zfs_vnodes_adjust_back();
1414 return (zfs_active_fs_count != 0);
1418 zfs_set_version(const char *name, uint64_t newvers)
1426 * XXX for now, require that the filesystem be unmounted. Would
1427 * be nice to find the zfsvfs_t and just update that if
1431 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
1434 error = dmu_objset_open(name, DMU_OST_ZFS, DS_MODE_OWNER, &os);
1438 error = zap_lookup(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
1442 if (newvers < curvers) {
1447 tx = dmu_tx_create(os);
1448 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, 0, ZPL_VERSION_STR);
1449 error = dmu_tx_assign(tx, TXG_WAIT);
1454 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 8, 1,
1457 spa_history_internal_log(LOG_DS_UPGRADE,
1458 dmu_objset_spa(os), tx, CRED(),
1459 "oldver=%llu newver=%llu dataset = %llu", curvers, newvers,
1464 dmu_objset_close(os);
1468 * Read a property stored within the master node.
1471 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
1477 * Look up the file system's value for the property. For the
1478 * version property, we look up a slightly different string.
1480 if (prop == ZFS_PROP_VERSION)
1481 pname = ZPL_VERSION_STR;
1483 pname = zfs_prop_to_name(prop);
1486 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
1488 if (error == ENOENT) {
1489 /* No value set, use the default value */
1491 case ZFS_PROP_VERSION:
1492 *value = ZPL_VERSION;
1494 case ZFS_PROP_NORMALIZE:
1495 case ZFS_PROP_UTF8ONLY:
1499 *value = ZFS_CASE_SENSITIVE;